A Parameter Identification Method for Permanent Magnet Synchronous Motors Based on PINN

Abstract

This invention relates to a parameter identification method for permanent magnet synchronous motors (PMSMs) based on PINN (Pin-Pin-N) architecture. The invention pertains to the field of PMSM drive control technology. It involves acquiring motor operating parameters, obtaining the motor's direct and quadrature axis voltages, currents, and electrical angular velocities; constructing a PMSM model; constructing a PINN model; and performing finite element simulation based on the constructed models to obtain motor operating data and identify the motor parameters. Compared to traditional PMSM models, this method offers higher accuracy and addresses the problems of reduced motor control performance, inaccurate dynamic response, and low efficiency caused by traditional motor equations under high loads or complex operating conditions, thus ensuring efficient and stable motor operation.

Description

Technical Field

[0001] This invention relates to the field of permanent magnet synchronous motor drive control technology, and is a parameter identification method for permanent magnet synchronous motors based on PINN. Background Technology

[0002] With the rapid popularization of electric vehicles, their safety and reliability technologies have received increasing attention and become an important research direction. Permanent magnet synchronous motors (PMSMs) have become the preferred power system for electric vehicles due to their advantages in power density and efficiency. However, in complex and ever-changing application environments, the accuracy and reliability of motor parameter identification face significant challenges. This not only affects the motor's control performance but also has a significant impact on the motor's condition monitoring system. Therefore, highly reliable and accurate motor parameter identification methods are crucial.

[0003] In practical embedded permanent magnet synchronous motors, phenomena such as magnetic saturation and cross-coupling exist, but traditional motor models do not take these into account. This affects the fidelity of the motor model and the control strategy. Currently, the processing of nonlinear parameters in motors mainly involves two methods for modeling: high-dimensional tables and fitting functions. Overall, while high-dimensional tables offer high modeling accuracy, the computation time and storage space required for interpolation increase significantly with the increase in table dimensions and number of tables. Although still significantly advantageous compared to finite element models, they cannot be directly used in embedded systems. In contrast, fitting functions have unique advantages in establishing motor models and have a broader prospect in practical applications.

[0004] Currently, parameter identification for permanent magnet synchronous motors (PMSMs) is mainly divided into offline and online identification. Online identification primarily includes recursive least squares, extended Kalma filtering, model reference adaptation, and some intelligent optimization algorithms. The main challenge in online identification lies in addressing the underrank problem of the equations. Common solutions include current injection or fixing two electrical parameters in the equations to their nominal values. Offline identification methods include particle swarm optimization, genetic algorithms, ant colony optimization, and neural networks. Compared to traditional intelligent optimization algorithms, neural networks, with their powerful data fitting capabilities and model flexibility, possess unique advantages in parameter identification. To date, the concept of Physically Constrained Neural Networks (PINNs) is gaining widespread attention. By incorporating physical information into the neural network, PINNs offer significant advantages in accuracy and robustness. Many scholars have already applied PINNs to electrical engineering, revealing their immense potential in parameter estimation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a parameter identification method for permanent magnet synchronous motors based on PINN (Permanent Magnet Synchronous Motor). This method exhibits high accuracy and robustness. First, a high-precision permanent magnet synchronous motor model is established using polynomial fitting. Then, a PINN motor parameter identification model driven by a hybrid mechanism and data approach is constructed. Simulations verify the applicability of this invention.

[0006] This invention provides the following technical solutions:

[0007] A parameter identification method for permanent magnet synchronous motors based on PINN, the method comprising the following steps:

[0008] Step 1: Collect motor operating parameters to obtain the motor's AC and DC axis voltage, current, and electric angular velocity;

[0009] Step 2: Build the PMSM model;

[0010] Step 3: Build the PINN model;

[0011] Step 4: Perform finite element simulation based on the built model to obtain motor operating data and identify motor parameters.

[0012] Preferably, step 2 specifically comprises:

[0013] The nonlinear voltage equation of the motor is expressed as:

[0014]

[0015] The relationship between inductance and magnetic flux in the motor voltage equation is as follows:

[0016]

[0017] Among them, L dd L qq L is the incremental inductance along the d-axis and q-axis. dq L qd The incremental mutual inductance between the d-axis and the q-axis;

[0018]

[0019] Polynomial fitting was performed on the direct and quadrature axis flux linkages and currents of the motor to determine the form of the flux linkage parameter equations and weighting coefficients, and the PMSM model was established as follows:

[0020]

[0021] Where a0, a1, a2, b0, c0, and c1 are polynomial coefficients.

[0022] Preferably, the PINN model includes a data layer and a physical layer:

[0023] The first half of the model is the data layer of the PINN model, which stores the motor's i... d i q u d u q ω e The sampling interval of the data is used as the input to the data layer, and the number of hidden layers and the number of neurons in each layer are configured according to the complexity of the data.

[0024] The input of the model is obtained by fitting the magnetic flux of the perpendicular axis through the neural network. The motor magnetic flux fitted by the neural network is also used as the input of the second half of the model. This layer is also the connection layer of the model.

[0025] Using the direct and quadrature axis flux linkages of the motor as input to the subsequent physical layer, the formula in the physical layer is as follows:

[0026]

[0027] After the output of the latter half of the physical layer, the predicted values ​​of the motor's direct and quadrature axis voltages are obtained. The mean square error between the predicted and actual values ​​is used as the learning error of the PINN model, as shown in the following formula:

[0028]

[0029] The learning error is propagated step by step through the physical layer and the data layer to update the learning gradient and the model's internal parameters.

[0030] Preferably, for the processing of differential terms, the backward Euler method is used to solve the problem and then the solution is substituted into the formula. In the configuration of the PINN model, the network structure is [6,100,100,100,100,2], where the numbers represent the number of neurons in each layer of the neural network.

[0031] Preferably, step 4 specifically comprises:

[0032] 200 consecutive sampling points were selected as the input to the PINN parameter identification model. Most of the data were the dynamic operating points of the motor, and a small portion were selected when the motor reached a steady state or all of them were selected as the dynamic operating points of the motor.

[0033] Preferably, the sampling frequency is set to 10kHz, the number of iterations is 200,000, and the learning rate is 0.1 in the simulation model.

[0034] Preferably, after testing, the parameter identification results obtained by the method are compared with simulation data. The identification results for parameters with large orders of magnitude are all within 8%, and the parameters a0 and R... s The error is within 3%.

[0035] A parameter identification system for permanent magnet synchronous motors based on PINN, the system comprising:

[0036] The data acquisition module collects motor operating parameters to obtain the motor's direct and quadrature axis voltage, current, and electric angular velocity.

[0037] PMSM model building module, which builds the PMSM model;

[0038] A PINN model building module, which builds the PINN model;

[0039] The identification module obtains motor operating data through finite element simulation based on the constructed model, and identifies the motor parameters.

[0040] A computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a PINN-based method for identifying parameters of a permanent magnet synchronous motor.

[0041] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement a PINN-based method for identifying parameters of a permanent magnet synchronous motor.

[0042] The present invention has the following beneficial effects:

[0043] Compared with the prior art, the present invention:

[0044] This invention first considers the influence of nonlinear factors such as magnetic saturation and cross-coupling in the motor, and establishes a novel physical model of the permanent magnet synchronous motor through polynomial fitting. Compared with the traditional permanent magnet synchronous motor model, it has higher accuracy and solves the problems of reduced motor control performance, inaccurate dynamic response, and low efficiency caused by traditional motor equations under high load or complex operating conditions, thus ensuring efficient and stable operation of the motor.

[0045] To address the challenges of complex operating conditions and accurate data acquisition in electric vehicles, this patent establishes the PINN motor parameter identification model by integrating a physical model of the motor into a neural network. Compared to traditional parameter identification methods, the PINN model offers the advantage of learning a more generalizable model with fewer data samples. Furthermore, when used for parameter identification, the PINN model yields results that conform to physical constraints, making the identification process more stable and reliable.

[0046] Finally, simulations verified the practicality of the proposed motor model and PINN parameter identification model, and the parameter identification accuracy also achieved the expected results. Attached Figure Description

[0047] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0048] Figure 1 The diagram shows the composition of the data layer and physical layer of this invention.

[0049] Figure 2 The diagram shows a PINN model driven by a hybrid mechanism and data of the present invention. Detailed Implementation

[0050] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] The present invention will be described in detail below with reference to specific embodiments. Specific Implementation Example 1:

[0053] according to Figures 1 to 2 As shown, the specific optimized technical solution adopted by the present invention to solve the above-mentioned technical problems is: The present invention relates to a parameter identification method for permanent magnet synchronous motors based on PINN.

[0054] A parameter identification method for permanent magnet synchronous motors based on PINN, the method comprising the following steps:

[0055] Step 1: Collect motor operating parameters to obtain the motor's AC and DC axis voltage, current, and electric angular velocity;

[0056] Step 2: Build the PMSM model;

[0057] Step 3: Build the PINN model;

[0058] Step 4: Perform finite element simulation based on the built model to obtain motor operating data and identify motor parameters. Specific Implementation Example 2:

[0060] The only difference between Embodiment 2 and Embodiment 1 of this application is that:

[0061] Step 2 specifically involves:

[0062] The nonlinear voltage equation of the motor is expressed as:

[0063]

[0064] The relationship between inductance and magnetic flux in the motor voltage equation is as follows:

[0065]

[0066] Among them, L dd L qq L is the incremental inductance along the d-axis and q-axis. dq L qd The incremental mutual inductance between the d-axis and the q-axis;

[0067]

[0068] Polynomial fitting was performed on the direct and quadrature axis flux linkages and currents of the motor to determine the form of the flux linkage parameter equations and weighting coefficients, and the PMSM model was established as follows:

[0069]

[0070] Where a0, a1, a2, b0, c0, and c1 are polynomial coefficients. Specific Implementation Example 3:

[0072] The only difference between Embodiment 3 and Embodiment 2 of this application is that:

[0073] The PINN model consists of a data layer and a physical layer:

[0074] The first half of the model is the data layer of the PINN model, which stores the motor's i... d i q u d u q ω e The sampling interval of the data is used as the input to the data layer, and the number of hidden layers and the number of neurons in each layer are configured according to the complexity of the data.

[0075] The input of the model is obtained by fitting the magnetic flux of the perpendicular axis through the neural network. The motor magnetic flux fitted by the neural network is also used as the input of the second half of the model. This layer is also the connection layer of the model.

[0076] Using the direct and quadrature axis flux linkages of the motor as input to the subsequent physical layer, the formula in the physical layer is as follows:

[0077]

[0078] After the output of the latter half of the physical layer, the predicted values ​​of the motor's direct and quadrature axis voltages are obtained. The mean square error between the predicted and actual values ​​is used as the learning error of the PINN model, as shown in the following formula:

[0079]

[0080] The learning error is propagated step by step through the physical layer and the data layer to update the learning gradient and the model's internal parameters. Specific Implementation Example 4:

[0082] The only difference between Embodiment 4 and Embodiment 3 of this application is that:

[0083] For the differential term, the backward Euler method is used to solve it and then the solution is substituted into the formula. In the configuration of the PINN model, the network structure is [6,100,100,100,100,2], where the numbers represent the number of neurons in each layer of the neural network. Specific Implementation Example 5:

[0085] The difference between Embodiment 5 and Embodiment 4 of the present invention lies only in:

[0086] Step 4 specifically involves:

[0087] 200 consecutive sampling points were selected as the input to the PINN parameter identification model. Most of the data were the dynamic operating points of the motor, and a small portion were selected when the motor reached a steady state or all of them were selected as the dynamic operating points of the motor. Specific Implementation Example Six:

[0089] The difference between Embodiment Six and Embodiment Five of the present invention lies only in:

[0090] In the simulation model, the sampling frequency was set to 10kHz, the number of iterations was 200,000, and the learning rate was 0.1. Specific Implementation Example 7:

[0092] The difference between Embodiment Seven and Embodiment Six of the present invention lies only in:

[0093] After testing, the method yielded parameter identification results. Comparison with simulation data showed that the identification rates for parameters of significant magnitude were all within 8%, particularly for parameters a0 and R. s The error is within 3%. Specific Implementation Example 8:

[0095] The difference between Embodiment 8 and Embodiment 7 of the present invention lies only in:

[0096] This invention provides a parameter identification system for permanent magnet synchronous motors based on PINN, the system comprising:

[0097] The data acquisition module collects motor operating parameters to obtain the motor's direct and quadrature axis voltage, current, and electric angular velocity.

[0098] PMSM model building module, which builds the PMSM model;

[0099] A PINN model building module, which builds the PINN model;

[0100] The identification module obtains motor operating data through finite element simulation based on the constructed model, and identifies the motor parameters. Specific Implementation Example Nine:

[0102] The difference between Embodiment Nine and Embodiment Eight of the present invention lies only in:

[0103] The present invention provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement a parameter identification method for a permanent magnet synchronous motor based on PINN. Specific Implementation Example 10:

[0105] The only difference between Embodiment 10 and Embodiment 9 of the present invention is that:

[0106] The present invention provides a computer device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement a parameter identification method for a permanent magnet synchronous motor based on PINN. Specific Implementation Example Eleven:

[0108] The only difference between Embodiment Eleven and Embodiment Ten of this invention is that:

[0109] To address the aforementioned technical problems, this invention provides a high-precision method for motor parameter identification, exhibiting strong accuracy and robustness. First, a high-precision permanent magnet synchronous motor model is established using polynomial fitting. Then, a PINN motor parameter identification model driven by a hybrid mechanism and data approach is constructed. Simulations verify the applicability of this invention.

[0110] Step 1: Data Acquisition: During simulation and experimentation, the dynamic operating conditions of the motor are first constructed to ensure sufficient data support for the subsequent parameter identification of the PINN model. During the simulation, motor operating parameters are collected to obtain the motor's direct and quadrature axis voltages, currents, and electrical angular velocities.

[0111] Step 2: Building the PMSM model: Traditional permanent magnet synchronous motor models do not take into account the influence of nonlinear factors such as magnetic saturation and cross-coupling. When the motor is running under high load, the control effect and state monitoring of the motor will be severely affected. Therefore, this invention proposes a new permanent magnet synchronous motor model established by polynomial fitting.

[0112] The traditional voltage equation for a permanent magnet synchronous motor is:

[0113]

[0114] In the formula u d u q i d i q R represents the voltage and current across the direct and quadrature axes of the motor. s L d L q ψ f For the stator resistance, quadrature-axis inductance, and permanent magnet flux linkage of the motor; ω e Let be the electric angular velocity of the motor.

[0115] The motor voltage equation considering motor nonlinearity can be expressed as:

[0116]

[0117] The relationship between inductance and magnetic flux in the motor voltage equation is as follows:

[0118]

[0119] In the formula L dd L qq L is the incremental inductance along the d-axis and q-axis. dq L qd This represents the incremental mutual inductance along the d-axis and q-axis. The direct-axis flux linkage of the motor has an even function relationship with the q-axis current, while the quadrature-axis flux linkage has an odd function relationship with the q-axis current. Simultaneously, L... dq L qd Let be the incremental mutual inductance of the d-axis and q-axis, and they are equal. Therefore, the equation is established as follows:

[0120]

[0121] Polynomial fitting was performed on the direct and quadrature axis flux linkages and currents of the motor to determine the form of the flux linkage parametric equations and the weighting coefficients, thus establishing the following parameterized model of the motor flux linkage:

[0122]

[0123] Where a0, a1, a2, b0, c0, and c1 are polynomial coefficients, and in this patent design, the equation for the direct-axis flux linkage of the motor contains a constant term. According to the physics of motors, the direct-axis flux linkage of a motor includes the flux linkage of the motor's permanent magnets, which is a value close to a constant. Therefore, in the constructed high-precision motor model, the constant term a0 can also represent the flux linkage of the motor's permanent magnets.

[0124] Step 3: Building the PINN model: Integrating knowledge of motor mechanisms into the neural network construction mechanism and data-driven PINN parameter identification model.

[0125] The PINN parameter identification model mainly consists of two parts: the data layer and the physical layer. Figure 1 As shown.

[0126] (1) The first part of the model is the data layer of the PINN model. First, the i of the motor is... d i q u d u q ω e The sampling interval of the data is used as the input to the data layer, and then the number of hidden layers and the number of neurons in each layer are configured according to the complexity of the data.

[0127] (2) The input of the model is obtained by fitting the direct-axis flux through the neural network. At the same time, the motor flux fitted by this part of the neural network is also used as the input of the second half of the model. Therefore, this layer is also the connection layer of the model.

[0128] (3) The direct and quadrature axis flux linkages of the motor are used as the input to the subsequent physical layer. The formulas in the physical layer are derived from formulas (2), (3), and (5), as shown in the following formula:

[0129]

[0130] (4) After the output of the second half of the physical layer, the predicted values ​​of the AC and DC axis voltages of the motor are obtained. The mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula:

[0131]

[0132] The final learning error can be propagated step-by-step through the physical and data layers to update the learning gradient and the model's internal parameters. This is a typical backpropagation learning method for training the parameters in the model. The parameters updated within the model mainly include the weights and biases between the neural networks in the data layers, as well as the parameters to be identified in the physical layers.

[0133] (5) In the formula, the differential term is solved by the backward Euler method and then substituted into the formula.

[0134] (6) Regarding the configuration of the PINN model, after multiple tests, its network structure was determined to be [6,100,100,100,100,2], where the numbers represent the number of neurons in each layer of the neural network.

[0135] Step 4: Simulation and Experimental Verification:

[0136] (1) This patent first performs finite element simulation on the motor to obtain the motor operating data. The parameters of the motor are shown in Table 1.

[0137] The parameters of the high-precision model of the motor are obtained by fitting the direct and quadrature axis flux linkages and currents of the motor according to formula (5), as shown in Table 2:

[0138] Table 1. Parameters of Permanent Magnet Synchronous Motor

[0139]

[0140] Table 2 Parameters of High-Precision Motor Model

[0141]

[0142] (2) Based on the obtained motor parameters, a dual closed-loop vector control model for the motor is constructed. The schematic diagram of the model is shown below. Figure 2 As shown.

[0143] In the simulation model settings, the motor operating conditions are set under various dynamic conditions for testing.

[0144] (3) Select 200 consecutive sampling points as input to the PINN parameter identification model. Most of the data are the dynamic operating points of the motor, and a small portion are selected when the motor reaches a steady state or all of them are selected as the dynamic operating points of the motor.

[0145] (4) This method ensures sufficient data to support the PINN model updates during parameter identification while minimizing data requirements compared to other models. The sampling frequency in the simulation model is set to 10kHz. The PINN model is configured with a network structure of [6, 100, 100, 100, 100, 2], where the numbers represent the number of neurons in each layer of the neural network. After multiple tests, the iteration count was determined to be 200,000 generations, with a learning rate of 0.1.

[0146] (5) Based on the data obtained from the simulation, the results obtained after parameter identification through the PINN model are shown in Table 3.

[0147] Table 3. Parameter Identification Results of Permanent Magnet Synchronous Motors

[0148]

[0149] After multiple tests, the parameter identification results were obtained. Comparison with simulation data showed that the identification rates for parameters with large orders of magnitude were all within 8%, and the parameters a0 and R... s The error is within 3%, which is a very good result. However, there is still much room for improvement in identifying results with a smaller order of magnitude.

[0150] The above description is merely a preferred embodiment of a PINN-based permanent magnet synchronous motor parameter identification method. The scope of protection for this PINN-based method is not limited to the above embodiments; all technical solutions falling within this conceptual framework are within the scope of protection of this invention. It should be noted that for those skilled in the art, any improvements and variations made without departing from the principles of this invention should also be considered within the scope of protection of this invention.

Claims

1. A PINN-based permanent magnet synchronous motor parameter identification method, characterized by: The method comprises the following steps: Step 1: Collect motor operating parameters to obtain motor direct-axis and quadrature-axis voltages, currents and electrical angular velocities; Step 2: Build a PMSM model; The step 2 is specifically: The motor voltage equation of the motor nonlinearity is expressed as: The relationship between inductance and flux in the motor voltage equation is: in, R s This is the stator resistance of the motor. i d 、i q This represents the current across the direct and quadrature axes of the motor. L dd 、L qq for d shaft and q Incremental inductance of the shaft, L dq 、L qd for d shaft and q Incremental mutual inductance of the shaft; The motor direct-axis and quadrature-axis fluxes and currents are polynomial fitted to determine the flux parameter equation form and weight coefficients, and the PMSM model is established as follows: wherein a 0 、a 1 、a 2 、b 0 、c 0 、c 1 is a polynomial coefficient; Step 3: Build a PINN model; Step 4: According to the built model, finite element simulation is performed to obtain motor operating data, and motor parameters are identified.

2. The method of claim 1 wherein the PI The NN model comprises a data layer and a physical layer: The first half of the model is the data layer of the PINN model, which inputs the motor's i d 、i q 、u d 、u q , ω e and the sampling interval of the data as the input of the data layer, the number of hidden layers and the number of neurons in each layer in the neural network are configured according to the complexity of the data. The input of the model is obtained through the fitting of the neural network to obtain the direct-axis and quadrature-axis fluxes, and the motor flux fitted by the neural network is also used as the input of the latter half of the model, and this layer is also the connection layer of the model; The motor direct-axis and quadrature-axis fluxes are used as the input of the subsequent physical layer, and the formula is constructed in the physical layer as shown in the following formula: After the output of the latter half of the physical layer, the predicted value of the motor direct-axis and quadrature-axis voltages is obtained, and the mean square error between the predicted value and the true value is used as the learning error of the PINN model as shown in the following formula: The learning error is transmitted through the physical layer and the data layer to update the learning gradient and the internal parameters of the model.

3. The method of claim 2, wherein: For the processing of the differential term, the backward Euler method is used to solve and then input into the formula, and in the configuration of the PINN model, the network structure is [6, 100, 100, 100, 100, 2], wherein the numbers represent the number of neurons in each layer of the neural network.

4. The method of claim 3, wherein: The step 4 is specifically: 200 continuous sampling points are selected as the input of the PINN parameter identification model, wherein most of the data are dynamic working points of the motor, and a small part of the data are selected as the motor steady-state or all are selected as the motor dynamic working points.

5. The method of claim 1, wherein: In the simulation model, the sampling frequency is set to 10kHZ, the iteration number is set to 200000 times, and the learning rate is set to 0.

1.

6. The method of claim 5, wherein: The method obtains parameter identification results after testing, and through comparison with simulation data, the parameter identification results of parameters of a large order of magnitude are all within 8%, and the errors of parameters of 0 and a 0 and R s are all within 3%.

7. A PINN-based permanent magnet synchronous motor parameter identification system, the system is based on the method of claim 1, characterized in that: The system comprises: A data acquisition module, which acquires motor operating parameters to obtain motor direct-axis and quadrature-axis voltages, currents and electrical angular velocities; A PMSM model building module, which builds a PMSM model; A PINN model building module, which builds a PINN model; An identification module, which performs finite element simulation according to the built model to obtain motor operating data and identifies motor parameters.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor to implement the method of any one of claims 1-6.

9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that: The processor executes the computer program to implement the method of any one of claims 1-6. The processor executes the computer program to implement the method of any one of claims 1-6.

Images